Characterization of Interactions Between Compounds and Polymers Using Pose Ensembles

This application claims priority to U.S. Provisional patent application No. 63/336,841, entitled “CHARACTERIZATION OF INTERACTIONS BETWEEN COMPOUNDS AND POLYMERS USING POSE ENSEMBLES,” filed Apr. 29, 2022, which is hereby incorporated by reference.

This application is directed to using models to characterize interactions between test compounds and target polymers.

Fundamentally, biological systems operate through the physical interaction of molecules, such as a compound with a target polymer. Structure-based, virtual high throughput screening (vHTS) methods have been used to characterize interactions between candidate (test) compounds and a target polymer through machine learning approaches. Such characterization, for instance, can report a continuous or categorical activity label, a PKa, or any other suitable metric to characterize the interaction between a candidate compound and a target polymer.

One drawback with vHTS machine learning methods is that they do not adequately take into account enthalpic and entropic components of receptor-ligand complex formation. Structure-based deep learning methods typically predict bioactivity from docked, static ligand poses. However, these approaches ignore the entropic contribution to the change in free energy. Predicting bioactivity from an ensemble of docked poses can overcome this limitation, but it requires that the model recognizes and is sensitive to different poses. However, conventional machine learning methods tend not to be sensitive to different poses.illustrates this insensitivity, where machine learning models such as convolutional neural networks incorrectly favor poses that have all the right components but are fundamentally incorrect overall.illustrates a situation in which the pose on the left and the pose on right have the same parts, two eyes, two eyebrows, a nose, lips, and the overall shape of a head. Teaching the machine learning model that the pose on the left, therefore, is the correct one can prove to be difficult. Because of this, there is an inherent pose insensitivity in conventional vHTS machine learning methods. This pose insensitivity can lead to the incorrect or inaccurate characterization of the interaction between a test compound and a target polymer. For instance, this pose insensitivity can lead a vHTS machine learning approach, which provides a categorical activity label for each compound in a screening library, to incorrectly label a certain percentage of the compounds in the screening library.

Given the above background, what is needed in the art are methods for imposing pose sensitivity on vHTS machine learning methods so that such vHTS machine learning methods are sensitive to bad poses.

The present disclosure addresses the problems identified in the background by making use of vHTS machine learning models that predict bioactivity from multiple poses concurrently. The disclosed vHTS machine learning models conditional multi-task architecture enforces sensitivity to distinct ligand poses, and its architecture includes an attention mechanism to exploit hidden correlations in pose distributions. The disclosed vHTS machine learning models improve bioactivity prediction compared to baseline models that predict from static ligand poses alone.

Accordingly, one aspect of the present disclosure provides a computer system for characterizing an interaction between a test compound and a target polymer. The computer system comprises one or more processors and memory addressable by the one or more processors. The memory stores at least one program for execution by the one or more processors. The at least one program comprises instructions for obtaining a plurality of sets of atomic coordinates. Each respective set of atomic coordinates in the plurality of sets of atomic coordinates comprises the test compound bound to the target polymer in a corresponding pose in a plurality of poses. In some embodiments, each respective set of atomic coordinates in the plurality of sets of atomic coordinates comprises atomic coordinates for at least 30 atoms.

For each respective set of atomic coordinates in the plurality of sets of atomic coordinates, the at least one program comprises instructions for inputting the respective set of atomic coordinates or an encoding of the respective set of atomic coordinates into a first neural network, to obtain a corresponding initial embedding as output of the first neural network, thereby obtaining a plurality of initial embeddings. Each initial embedding in the plurality of initial embeddings corresponds to a set of atomic coordinates in the plurality of sets of atomic coordinates. The first neural network comprises more than 400 parameters.

The at least one program further comprises instructions for applying an attention mechanism to the plurality of initial embeddings, in concatenated form, thereby obtaining an attention embedding.

The at least one program further comprises instructions for applying a pooling function to the attention embedding to derive a pooled embedding.

The at least one program further comprises instructions for inputting the pooled embedding into a first model thereby obtaining a first interaction score of an interaction between the test compound and the target polymer. The first model comprises more than 400 parameters.

In some embodiments, the first interaction score represents a binding coefficient of the test compound to the target polymer. In some such embodiments the binding coefficient is an IC, EC, Kd, KI, or pKI for the test compound with respect to the target polymer.

In some embodiments, the first interaction score represents an in silico pose quality score of the test compound to the target polymer.

In some embodiments, the first model is a fully connected second neural network.

In some embodiments, the at least one program further comprises instructions for inputting the pooled embedding into a second model thereby obtaining a second interaction score of an interaction between the test compound and the target polymer. In such embodiments, the first model is a first fully connected neural network, the second model is a second fully connected neural network, the first interaction score represents an in silico pose quality score of the test compound to the target polymer, and the second interaction score represents an in silico pose quality score of the test compound to the target polymer. In some such embodiments, the at least one program further comprises instructions for inputting the first interaction score and the second interaction score into a third model to obtain a third interaction score, where the third model is a third fully connected neural network. In some such embodiments, the third interaction score is discrete-binary activity score with a first value when the test compound is determined by third model to be inactive and a second value when the test compound is determined by the third model to be active.

In some embodiments, the target polymer is a protein, a polypeptide, a polynucleic acid, a polyribonucleic acid, a polysaccharide, or an assembly of any combination thereof.

In some embodiments, each set of atomic coordinates in the plurality of sets of atomic coordinates comprises three-dimensional coordinates {x, . . . , x} for at least a portion of the polymer from a crystal structure of the target polymer resolved at a resolution of 2.5 Å or better or a resolution of 3.3 Å or better.

In some embodiments, each set of atomic coordinates in the plurality of sets of atomic coordinates comprises an ensemble of three-dimensional coordinates for at least a portion of the target polymer determined by nuclear magnetic resonance, neutron diffraction, or cryo-electron microscopy.

In some embodiments, the first interaction score is a binary score, where a first value for the binary score represents an IC, EC, Kd, KI, or pKI for the test compound with respect to the target polymer that is above a first threshold, and a second value for the binary score represents an IC, EC, Kd, KI, or pKI for the test compound with respect to the target polymer that is below the first threshold.

In some embodiments, the test compound satisfies two or more rules, three or more rules, or all four rules of the Lipinski's rule of Five: (i) not more than five hydrogen bond donors, (ii) not more than ten hydrogen bond acceptors, (iii) a molecular weight under 500 Daltons, and (iv) a Log P under 5.

In some embodiments, the test compound is an organic compound having a molecular weight of less than 500 Daltons, less than 1000 Daltons, less than 2000 Daltons, less than 4000 Daltons, less than 6000 Daltons, less than 8000 Daltons, less than 10000 Daltons, or less than 20000 Daltons.

In some embodiments, the test compound is an organic compound having a molecular weight of between 400 Daltons and 10000 Daltons.

In some embodiments, the plurality of sets of atomic coordinates consists of between 3 and 64 poses. In some embodiments, the plurality of sets of atomic coordinates consists of between 2 and 64 poses.

In some embodiments, the first neural network is a convolutional neural network. In some such embodiments, each respective set of atomic coordinates in the plurality of sets of atomic coordinates comprises atomic coordinates for at least 80 atoms.

In some embodiments, the first neural network is a graph neural network. In some such embodiments, the graph neural network is characterized by an initial embedding layer and a plurality of interaction layers that each contribute an interaction data structure, in a plurality of interaction data structures, for each atom in the respective set of atomic set of atomic coordinates for the corresponding pose in the plurality of poses that are pooled to form the corresponding initial embedding for the corresponding pose.

In some embodiments, the first neural network is a equivariant neural network or a message passing neural network.

In some embodiments, the first neural network comprises a plurality of graph convolutional blocks and each block considers connectivity within the respective set of atomic coordinates using a plurality of radial graphs.

In some embodiments, the first neural network comprises 1×10parameters.

In some embodiments, the corresponding initial embedding comprises a data structure having between 128 and 768 values. In some embodiments, the corresponding initial embedding comprises a data structure having more than 100 values. In some embodiments, the corresponding initial embedding comprises a data structure having more than 80 values, more than 100 values, more than 120 values, more than 140 values, or more than 160 values. In some embodiments, the corresponding initial embedding comprises a data structure consisting of between 100 values and 2000 values.

In some embodiments, the plurality of initial embeddings comprises a first plurality of values, and the applying the attention mechanism comprises: (i) inputting the first plurality of values into an attention neural network thereby obtaining a first plurality of weights, where each weight in the first plurality of weights corresponds to a respective value in the first plurality of values, and (ii) weighting each respective value in the first plurality of values by the corresponding weight in the plurality of weights thereby obtaining the attention embedding. In some such embodiments, the first plurality of weights sum to one and each weight in the first plurality of weights is a scalar value between zero and one.

In some embodiments, the pooling function collapses the attention embedding into the pooled embedding by applying a statistical function to combine each portion of the attention embedding representing a different pose in the plurality of poses to form the pooled embedding. In some such embodiments, the attention embedding includes a corresponding plurality of values for a corresponding plurality of elements for each respective pose in the plurality of poses and the statistical function is a maximum function that takes a maximum value across corresponding elements of each respective pose represented in the attention embedding to form the pooled embedding. In some embodiments, the attention embedding includes a corresponding plurality of values for a corresponding plurality of elements for each respective pose in the plurality of poses and the statistical function is an average function that averages the corresponding elements of each respective pose represented in the attention embedding to form the pooled embedding.

In some embodiments, the first model is a regression task and the first interaction score quantifies the interaction between the test compound and the target polymer.

In some embodiments, the first model is a classification task and the first interaction score classifies the interaction between the test compound and the target polymer.

Another aspect of the present disclosure provides a method for characterizing an interaction between a test compound and a target polymer. The method comprises obtaining a plurality of sets of atomic coordinates. Each respective set of atomic coordinates in the plurality of sets of atomic coordinates comprises the test compound bound to the target polymer in a corresponding pose in a plurality of poses. In some embodiments, each respective set of atomic coordinates in the plurality of sets of atomic coordinates comprises atomic coordinates for at least 30 atoms. Further in the method, for each respective set of atomic coordinates in the plurality of sets of atomic coordinates, the respective set of atomic coordinates, or an encoding of the respective set of atomic coordinates, is inputted into a first neural network to obtain a corresponding initial embedding as output of the first neural network. In this way a plurality of initial embeddings is obtained, where each initial embedding in the plurality of initial embeddings corresponds to a set of atomic coordinates in the plurality of sets of atomic coordinates. In some such embodiments the first neural network comprises more than 400 parameters. The method further comprises applying an attention mechanism to the plurality of initial embeddings, in concatenated form, thereby obtaining an attention embedding. The method further comprises applying a pooling function to the attention embedding to derive a pooled embedding. The method further comprises inputting the pooled embedding into a first model thereby obtaining a first interaction score of an interaction between the test compound and the target polymer. In some embodiments the first model comprises more than 400 parameters.

Another aspect of the present disclosure provides a non-transitory computer readable storage medium. The non-transitory computer readable storage medium stores instructions, which when executed by a computer system, cause the computer system to perform a method for characterizing an interaction between a test compound and a target polymer. The method comprises obtaining a plurality of sets of atomic coordinates. each respective set of atomic coordinates in the plurality of sets of atomic coordinates comprises the test compound bound to the target polymer in a corresponding pose in a plurality of poses. In some embodiments, each respective set of atomic coordinates in the plurality of sets of atomic coordinates comprises atomic coordinates for at least 30 atoms. The method further comprises, for each respective set of atomic coordinates in the plurality of sets of atomic coordinates, inputting the respective set of atomic coordinates or an encoding of the respective set of atomic coordinates into a first neural network to obtain a corresponding initial embedding as output of the first neural network. In this way a plurality of initial embeddings is obtained where each initial embedding in the plurality of initial embeddings corresponds to a set of atomic coordinates in the plurality of sets of atomic coordinates. In some embodiments the first neural network comprises more than 400 parameters. The method further comprises applying an attention mechanism to the plurality of initial embeddings, in concatenated form, thereby obtaining an attention embedding. The method further comprises applying a pooling function to the attention embedding to derive a pooled embedding. The method further comprises inputting the pooled embedding into a first model thereby obtaining a first interaction score of an interaction between the test compound and the target polymer. In some embodiments, the first model comprises more than 400 parameters.

Like reference numerals refer to corresponding parts throughout the several views of the drawings.

Reference will now be made in detail to embodiments, examples of which are illustrated in the accompanying drawings. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. However, it will be apparent to one of ordinary skill in the art that the present disclosure may be practiced without these specific details. In other instances, well-known methods, procedures, components, circuits, and networks have not been described in detail so as not to unnecessarily obscure aspects of the embodiments.

The present disclosure provides systems and methods for characterizing an interaction between a compound and a polymer. In the methods a plurality of sets of atomic coordinates are obtained. Each of these sets of atomic coordinates comprises the compound bound to the polymer in a corresponding pose in a plurality of poses. Each respective set of atomic coordinates, or an encoding thereof, is sequentially inputted into a neural network to obtain a corresponding initial embedding as output. In this way a plurality of initial embeddings is calculated. Each initial embedding corresponds to a set of atomic coordinates in the plurality of sets of atomic coordinates. An attention mechanism is applied to the plurality of initial embeddings, in concatenated form, to obtain an attention embedding. In some embodiments the attention mechanism is a neural network that is trained on test data to emphasize some portions of the plurality of initial embeddings while deemphasizing some portions of the plurality of initial embeddings. A pooling function is applied to the attention embedding to derive a pooled embedding. Thus, the pooling function collapses all the initial embeddings representing the plurality of poses into a single composite embedding that represents all the poses in the plurality of poses. The pooled embedding is inputted into a model to obtain an interaction score of the interaction between the compound and the polymer.

It will also be understood that, although the terms first, second, etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first subject could be termed a second subject, and, similarly, a second subject could be termed a first subject, without departing from the scope of the present disclosure. The first subject and the second subject are both subjects, but they are not the same subject.

The terminology used in the present disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the description of the invention and the appended claims, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term “and/or” as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items. It will be further understood that the terms “comprises” and/or “comprising,” when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

As used herein, the term “if” may be construed to mean “when” or “upon” or “in response to determining” or “in response to detecting,” depending on the context. Similarly, the phrase “if it is determined” or “if [a stated condition or event] is detected” may be construed to mean “upon determining” or “in response to determining” or “upon detecting [the stated condition or event]” or “in response to detecting [the stated condition or event],” depending on the context.

illustrates a computer systemfor characterizing an interaction between a test compound and a target polymer. For instance, it can be used as a binding affinity prediction system to generate accurate predictions regarding the binding affinity of one or more test compounds with a target polymer.

Referring to, in typical embodiments, computer systemcomprises one or more computers. For purposes of illustration in, the computer systemis represented as a single computer that includes all of the functionality of the disclosed computer system. However, the present disclosure is not so limited. The functionality of the computer systemmay be spread across any number of networked computers and/or reside on each of several networked computers and/or virtual machines. One of skill in the art will appreciate that a wide array of different computer topologies are possible for the computer systemand all such topologies are within the scope of the present disclosure.

Turning towith the foregoing in mind, the computer systemcomprises one or more processing units (CPUs), a network or other communications interface, a user interface(e.g., including an optional displayand optional keyboardor other form of input device), a memory(e.g., random access memory, persistent memory, or combination thereof), one or more magnetic disk storage and/or persistent devicesoptionally accessed by one or more controllers, one or more communication bussesfor interconnecting the aforementioned components, and a power supplyfor powering the aforementioned components. To the extent that components of memoryare not persistent, data in memorycan be seamlessly shared with non-volatile memoryor portions of memorythat are non-volatile/persistent using known computing techniques such as caching. Memoryand/or memorycan include mass storage that is remotely located with respect to the central processing unit(s). In other words, some data stored in memoryand/or memorymay in fact be hosted on computers that are external to computer systembut that can be electronically accessed by the computer systemover an Internet, intranet, or other form of network or electronic cable using network interface. In some embodiments, the computer systemmakes use of models that are run from the memory associated with one or more graphical processing units in order to improve the speed and performance of the system. In some alternative embodiments, the computer systemmakes use of models that are run from memoryrather than memory associated with a graphical processing unit.

The memoryof the computer systemstores:

In some implementations, one or more of the above identified data elements or modules of the computer systemare stored in one or more of the previously mentioned memory devices, and correspond to a set of instructions for performing a function described above. The above identified data, modules or programs (e.g., sets of instructions) need not be implemented as separate software programs, procedures or modules, and thus various subsets of these modules may be combined or otherwise re-arranged in various implementations. In some implementations, the memoryand/or(and optionally) optionally stores a subset of the modules and data structures identified above. Furthermore, in some embodiments the memoryand/or(and optionally) stores additional modules and data structures not described above. In some embodiments, the first neural networkis replaced with another form of model.

Now that a system for characterizing an interaction between a test compound and a target polymer has been disclosed, methods for performing such characterization is detailed with reference toand discussed below.

Block. Referring to blockof, a computer systemfor characterizing an interaction between a test compound and a target polymer is provided. As discussed above in conjunction with, the computer systemcomprises one or more processorsand memory/addressable by the one or more processors. The memory stores at least one program for execution by the one or more processors. The at least one program comprises instructions detailed below.

Blocksthrough. Referring to blockof, a plurality of sets of atomic coordinates is obtained. Each respective set of atomic coordinatesin the plurality of sets of atomic coordinates comprises the test compound bound to the target polymer in a corresponding posein a plurality of poses. In other words, with reference to, each respective set of atomic coordinatesincludes both the atomic coordinates of the test compound and at least a subset of the spatial coordinates of the target polymer that is considered by the first neural network. For instance in some embodiments, the set of atomic coordinatesconsists of the atomic coordinates of the test compound and the atomic coordinates of the portion of the target polymer that make up an active site to which the test compound has been docked. In some embodiments the target polymer comprises multiple active sites, and the test compound has been docked to one of the active sites.illustrates a poseof a test compound in an active site of a target polymer. Each respective set of atomic coordinates in the plurality of sets of atomic coordinates comprises atomic coordinates for at least 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95, 100, 150, 200, 250, or 300 atoms. In some embodiments, a set of atomic coordinates in the plurality of sets of atomic coordinates comprises atomic coordinates for at least 400 atoms of the target polymer in addition to the atomic coordinates of the test compound. In some embodiments, a set of atomic coordinates in the plurality of sets of atomic coordinates comprises atomic coordinates for at least 25 atoms, at least 50 atoms, at least 100 atoms, at least 200 atoms, at least 300 atoms, at least 400 atoms, at least 1000 atoms, at least 2000 atoms, or at least 5000 atoms of the target polymer in addition to the atomic coordinates of the test compound. In some embodiments, only the coordinates of the active site of the target polymerwhere ligands are expected to bind the target polymer is present in each respective set of atomic coordinates in the plurality of sets of atomic coordinates in addition to the coordinates of the corresponding test compounds.

Referring to blockof, in some embodiments, the target polymer is a protein, a polypeptide, a polynucleic acid, a polyribonucleic acid, a polysaccharide, or an assembly of any combination thereof.